Institution profile

University of Warwick

Academic institutioneurope · gb
Official website
Research library563linked papers
Opportunities0open roles
Selected work

Representative Papers

SIGACT News Complexity Theory Column 124 Meta-Mathematics of Computational Complexity Theory

Mar 24, 2025ACM SIGACT News

This paper addresses fundamental open problems in computational complexity—such as P vs NP and the nonexistence of polynomial-size circuits for SAT—within weak formal systems like bounded arithmetic $S^1_2$. Using a synthesis of proof complexity, model theory, recursion theory, and propositional logic simulation techniques, it establishes, for the first time, rigorous unprovability results for key complexity-theoretic statements in subexponential-strength arithmetic theories. The main contributions are: (1) proving that assertions such as “SAT has no polynomial-size circuits” are independent of $S^1_2$; (2) establishing a tight correspondence between proof complexity lower bounds and circuit lower bounds; and (3) exposing deep metatheoretic barriers preventing any feasible formal proof of P = NP, thereby offering a novel logical foundation for complexity theory.

91 citations5 influentialRead paper

Attention-Based Adversarial Robust Distillation in Radio Signal Classifications for Low-Power IoT Devices

Feb 01, 2023IEEE Internet of Things Journal

Lightweight Transformers for wireless modulation classification on IoT devices suffer from vulnerability to adversarial attacks—particularly cross-architecture transfer attacks—while struggling to balance robustness and inference efficiency. Method: This paper proposes an attention-mechanism-transfer-based adversarial robust knowledge distillation framework. It distills robust attention maps learned by a large teacher model during adversarial training into a compact Transformer student, without incurring additional inference overhead. The approach integrates adversarial training, attention-guided knowledge distillation, and lightweight architecture design. Contribution/Results: Under FGSM and PGD white-box attacks, the distilled lightweight model achieves significantly enhanced robustness and effectively mitigates adversarial sample transfer across architectures. Experiments demonstrate high-accuracy, robust modulation classification under resource constraints—retaining low latency and power consumption—thus overcoming the critical bottleneck of co-optimizing model lightness and security.

7 citationsRead paper

Rate Optimality and Phase Transition for User-Level Local Differential Privacy

May 20, 2024

This paper investigates the fundamental statistical estimation limits under user-level local differential privacy (LDP) in the multi-observation setting: each of $n$ users holds $T$ independent observations. The authors establish the first general information-theoretic lower bound for user-level LDP, revealing a $T$-driven phase transition in both mean estimation and nonparametric density estimation—namely, a critical threshold exists below which estimation risk does not vanish with $n$, and above which consistent estimation becomes possible. They further demonstrate that high-dimensional sparse mean estimation is feasible under user-level LDP but impossible under standard item-level LDP. Tight (up to logarithmic factors) minimax upper and lower bounds are derived for univariate/multivariate mean estimation, sparse mean estimation, and density estimation, explicitly characterizing the critical interplay among $T$, dimension $d$, and sparsity $s$. These results provide novel feasibility criteria for statistical inference under user-level privacy constraints.

4 citations1 influentialRead paper

Satisfactory Medical Consultation based on Terminology-Enhanced Information Retrieval and Emotional In-Context Learning

Mar 22, 2025

Current large language models (LLMs) struggle to simultaneously achieve clinical-level domain expertise, sustained multi-turn interaction, and empathetic responsiveness—key requirements for high clinical satisfaction in medical consultation. To address this, we propose a clinical-satisfaction-oriented medical consultation framework featuring two novel components: (1) a terminology-driven implicit reasoning retrieval mechanism that enhances domain accuracy, and (2) an emotion-attribute memory model trained on unlabeled conversational corpora to capture affective patterns. The framework integrates Terminology-Enhanced Implicit Retrieval (TEIR) and Emotion-Informed Context Learning (EICL), enabling proactive symptom elicitation and empathetic, context-aware multi-turn dialogue. Evaluated on over 800,000 real-world Chinese doctor–patient conversations, our method significantly extends the effective context window of LLMs and consistently outperforms five strong baselines across BLEU, ROUGE, and other standard metrics. Empirical results further demonstrate measurable improvements in patient satisfaction scores.

2 citationsRead paper

Hamiltonian Property Testing

Mar 05, 2024arXiv.org

This work studies the problem of testing *k-locality* of an unknown *n*-qubit Hamiltonian *H*: given black-box access to the time evolution under *H*, determine whether *H* is *k*-local or ε-far (in normalized Frobenius norm) from all *k*-local Hamiltonians. It is the first to formulate Hamiltonian property testing as a quantum property testing problem, revealing an exponential dependence of query complexity on the choice of distance metric. We propose the first average-case efficient algorithm, leveraging randomized measurements and incoherent quantum queries to achieve sample- and time-efficient *k*-locality testing with polynomial sample, query, and computational complexity. Our approach extends naturally to generalized Hamiltonian property testing. Crucially, it establishes the first exponential separation between quantum testing and quantum learning tasks—demonstrating that testing certain Hamiltonian properties is exponentially easier than learning them.

2 citationsRead paper
Recent publications

Latest Papers

Mined from Scientific Literature: Process Schemas for Atomic Layer Deposition and Etching in Materials Science

Sep 10, 2026

"This study addresses the challenges in data comparison and reutilization within the field of materials science, specifically due to the inconsistent reporting formats of atomic layer deposition (ALD) and atomic layer etching (ALE) experiments and simulations. The work proposes and expertly reviews four JSON schemas based on the QUDT standard, designed to standardize the description of materials, process conditions, configurations, and results in ALD/ALE processes, ensuring data consistency and interoperability. The innovation lies in the development and application of these specialized JSON schemas, which, through the use of a schema-miner toolset, enable effective extraction and structured representation of literature content, thereby promoting knowledge sharing. Additionally, the study provides a comparative analysis of the schemas and publishes related structured records on the ORKG platform, demonstrating their potential for guiding information extraction tasks."

0 citationsRead paper